Haokun Qin, Cofounder @ Gale (YC W25). We make immigration simple.
Some of the most transformative applications of AI operate far away from the spotlights of Silicon Valley, seen only by back-office workers who are now saving thousands of hours in rote menial work.
Unsurprisingly, then, even the U.S. Citizenship and Immigration Services (USCIS) has started to leverage various forms of AI to introduce some much-needed efficiency into an archaic system.
It used to take an average of 10.5 months to process a case. After implementing AI, it now takes an average of 6.1 months per case, allowing for the USCIS to process an unprecedented 10.9 million cases last year, which reduced their backlog by 15% for the first time in over a decade.
Removing a human from the loop doesn’t come without consequences, however, and understanding exactly what AI is doing here is key to understanding modern immigration practices and the future of this industry.
A New Era In Document Processing
In 2020, the USCIS rolled out an Evidence Classifier System, which classifies and tags the evidence in an immigration case into several common categories (e.g., passport pages, marriage certificate, etc.), making it easier to assess evidence. This relatively simple system has saved USCIS workers 24 million page scrolls in two years since its inception, making it hard to believe that a human used to do this manually.
The technology behind this is likely a combination of a computer vision backbone, combined with a classification head, where the input is a page in a PDF, and the output is the type of document. It works by applying convolutional neural networks (CNNs) and optical character recognition (OCR) to extract information from each page before using that information to classify them into their respective categories.
What does this mean for the applicant? Faster processing times, but nonstandard documents may be harder to categorize.
Enhanced Officer Training With LLMs
In 2024, the USCIS piloted an Officer Training Tool, which allowed officers to practice interviews for refugee/asylum seeker cases with conversational AI. The unprecedented realism of large language models (LLMs) introduced a new paradigm of training possibilities, ushering in a world where officers can have an immersive interview experience without the need for a second person.
It’s likely that the tool is based on a transformer-based LLM, fine-tuned on immigration law, case precedents and policy guidelines, then fed realistic refugee/asylum cases. It’s still very early into the program, but optimistically, this means that USCIS officers with whom the public interacts in the future will be more knowledgeable and better equipped.
AI Fraud Detection
Perhaps the most critical and contentious aspect of AI in USCIS is its use in fraud detection. In recent years, USCIS has responded to growing immigration fraud with several countermeasures, including the deployment of sophisticated AI models to flag anomalies in applications. It’s possible that they could be using unsupervised learning techniques like the Isolation Forest algorithm to pinpoint unusual patterns in the data that may suggest fraudulent activity.
It’s also likely that identity matching algorithms—AI that can assign a measure of case similarity—are used to flag applications that may be fraudulent duplicates. So, while these tools significantly reduce fraud and increase general efficiency, they can also be very unforgiving to genuine applicants. Even minor discrepancies, such as slight differences in how your name is recorded across documents, might trigger additional scrutiny.
What This Means For Business Leaders
On the bright side, we are indisputably entering an exciting new frontier of unparalleled efficiency for immigration and the world at large. However, for anyone applying for a visa, green card or naturalization, the message is clear: Every name, date and document must be consistent and error-free, given that even minor mistakes can incur additional requests for evidence (RFEs).
Companies can prepare HR and legal teams for AI-driven immigration by standardizing documents, ensuring clean data and drawing lessons from legal and financial automation. Even USCIS notes that moving from paper to standardized digital filings yields immediate cost and efficiency gains, and studies find AI can boost efficiency by 40% and cut costs by 30%.
Leaders should caution visa-seekers that what humans once handled is now enforced by rigid automation—every name and date must exactly match, and documents must be in standard formats with clear scans and certified translations to avoid automated RFEs.
Internally, firms can mirror these strategies by deploying robotic process automation (RPA) bots (e.g., UiPath, Automation Anywhere or Blue Prism) for onboarding or compliance, contract AI (e.g., Kira Systems, Luminance or Ironclad) for legal or GPT-driven agents for customer onboarding and support. Case studies show contract reviews 80% faster and routine queries automated to save thousands of hours, freeing teams for higher-value work.
I sometimes wonder where notably archaic industries like immigration will head in 20 years. My personal belief is that the demand for immigration technology will surge; highly educated lawyers shouldn’t be filing paperwork all day and applicants deserve to have fast turnaround times. I’m excited to be part of this change.
Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?
Read the full article here
